ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Christian's kokkos patch that allows concurrent host and device threading

Browse Source 
git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@14752 f3b2605a-c512-4ea7-a41b-209d697bcdaa
This commit is contained in:
stamoor
2016-03-18 14:50:28 +00:00
parent b760ca8038
commit 042f50ac9c
81 changed files with 1282 additions and 527 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
src/KOKKOS/atom_vec_atomic_kokkos.cpp
View File

@ -896,6 +896,7 @@ void AtomVecAtomicKokkos::unpack_border_vel(int n, int first, double *buf)
last = first + n;
for (i = first; i < last; i++) {
if (i == nmax) grow(0);
modified(Host,X_MASK|V_MASK|TAG_MASK|TYPE_MASK|MASK_MASK);
h_x(i,0) = buf[m++];
h_x(i,1) = buf[m++];
h_x(i,2) = buf[m++];
@ -1377,6 +1378,43 @@ void AtomVecAtomicKokkos::sync(ExecutionSpace space, unsigned int mask)
/* ---------------------------------------------------------------------- */
void AtomVecAtomicKokkos::sync_overlapping_device(ExecutionSpace space, unsigned int mask)
{
if (space == Device) {
if ((mask & X_MASK) && atomKK->k_x.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_x_array>(atomKK->k_x,space);
if ((mask & V_MASK) && atomKK->k_v.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_v_array>(atomKK->k_v,space);
if ((mask & F_MASK) && atomKK->k_f.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_f_array>(atomKK->k_f,space);
if ((mask & TAG_MASK) && atomKK->k_tag.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_tagint_1d>(atomKK->k_tag,space);
if ((mask & TYPE_MASK) && atomKK->k_type.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_int_1d>(atomKK->k_type,space);
if ((mask & MASK_MASK) && atomKK->k_mask.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_int_1d>(atomKK->k_mask,space);
if ((mask & IMAGE_MASK) && atomKK->k_image.need_sync<LMPDeviceType>())
perform_async_copy<DAT::tdual_imageint_1d>(atomKK->k_image,space);
} else {
if ((mask & X_MASK) && atomKK->k_x.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_x_array>(atomKK->k_x,space);
if ((mask & V_MASK) && atomKK->k_v.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_v_array>(atomKK->k_v,space);
if ((mask & F_MASK) && atomKK->k_f.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_f_array>(atomKK->k_f,space);
if ((mask & TAG_MASK) && atomKK->k_tag.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_tagint_1d>(atomKK->k_tag,space);
if ((mask & TYPE_MASK) && atomKK->k_type.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_int_1d>(atomKK->k_type,space);
if ((mask & MASK_MASK) && atomKK->k_mask.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_int_1d>(atomKK->k_mask,space);
if ((mask & IMAGE_MASK) && atomKK->k_image.need_sync<LMPHostType>())
perform_async_copy<DAT::tdual_imageint_1d>(atomKK->k_image,space);
}
}
/* ---------------------------------------------------------------------- */
void AtomVecAtomicKokkos::modified(ExecutionSpace space, unsigned int mask)
{
if (space == Device) {
@ -1397,3 +1435,4 @@ void AtomVecAtomicKokkos::modified(ExecutionSpace space, unsigned int mask)
if (mask & IMAGE_MASK) atomKK->k_image.modify<LMPHostType>();
}
}